The Pulse
Embed Security lets customers create AI investigation agents
Embed Security has introduced customer-built investigation agents that analysts can create by describing their methods in natural language. The company also expanded its organizational context feature and added recommendations for improving customer-authored detections.

Security teams can now describe their investigation methods in natural language to create agents that run within Embed Security’s AI security operations platform. Embed Security announced the customer-built agents on November 11, alongside expanded organizational context and a feature for recommending changes to detections.
Agents from analyst methods
The iSteps addition extends an existing library of investigation modules. The existing library combines deterministic logic with large language model reasoning and was developed by the company’s security and AI specialists.
According to the company, analysts describe how they approach a security question, and the platform turns that expertise into investigations that run autonomously when relevant. The customer-built agents use the same orchestration framework as the existing library, without requiring customers to program AI workflows.
In its platform FAQ, Embed describes an iStep as a unit that gathers evidence, answers a specific investigative question and contributes to a conclusion. The company says analysts can see the steps used to reach a decision.
Context that can change
NoiseIQ Memories expands the platform’s existing organizational context capabilities, giving teams another way to provide information about their environment. Rather than relying solely on predefined relationships, the platform connects organizational knowledge with investigative evidence, including context that changes over time, according to Embed.
The company describes NoiseIQ more broadly as a way to capture operational knowledge and apply it during investigations. Its FAQ says that knowledge can include an organization’s users, assets, policies and historical activity.
Improving detections
The new Detection Engineering capability uses investigation outcomes to help teams identify and improve detections across endpoint, cloud, email and identity. By analyzing investigated alerts, it provides a view of detection performance and recommends improvements to customer-authored detections in supported security information and event management platforms, the company says.
Jeffrey Johns, Embed Security’s chief technology officer, said the feature uses investigation outcomes to produce information teams can apply to the detections themselves. The recommendations are described as ready to apply.
How investigations fit into the platform
The broader platform connects to existing security tools through application programming interfaces, according to Embed’s FAQ. Its listed data sources include endpoint detection and response, email security tools, cloud environments, identity systems, threat intelligence, historical data and organizational context.
Embed says an investigation includes a classification of benign, malicious or inconclusive, along with evidence, a summary, reasoning and suggested next steps. If it cannot confidently classify an alert as benign or malicious, the investigation is escalated as inconclusive for analyst review.
Organizations retain control of remediation decisions, the company says. Optional integrations with security orchestration, automation and response platforms can enable automated actions where appropriate and under customer-defined governance.
Sources
Related stories
- Amazon says autonomous AI agents wrote its AWS Emulator
- AWS introduces cross-framework evaluation system for AI agents
- ElevenLabs previews enterprise agents and model fine-tuning
- Anthropic opens self-hosted Claude Managed Agents to private beta
- Entire gives AI agents service accounts for repository access